Repolode
FlashML-org

FlashML-org/FreeVideo

Local video generation app and ComfyUI plugin that runs MiniMax H3 on consumer GPUs with as little as 8 GB of VRAM.

Stars
2,814
Forks
338
Open issues
416
Last commit
15 hours ago

What is it?

FreeVideo is a local inference engine for the MiniMax H3 video model, built on the OpenVDN 8-step VDN-H3 model. It targets users who want to generate videos on their own consumer GPU rather than through a hosted service. It coordinates VRAM, system memory and disk by streaming weights, prefetching them asynchronously and computing in chunks, so generation can run with as little as 8 GB of VRAM and 16 GB of RAM.

The project ships as a ComfyUI plugin with a dedicated creative workspace, plus launchers for Windows, macOS (Apple silicon preview) and Linux. Launchers install ComfyUI, the runtime and the models, and can reuse existing model folders or install offline. Inputs include text prompts, first and last frames, and image, video and audio references, and users can choose one of four quality levels per video.

Key features

  • 01Chooses int8 or FP8 weight formats based on the GPU.
  • 02Weight streaming and chunked computation keep peak memory low.
  • 03Accepts text prompts, first and last frames, and image, video and audio references.
  • 04Works with community MiniMax H3 LoRAs inside ComfyUI.
  • 05Offers a creative workspace and a node view for custom workflows.
  • 06Launchers handle installation of ComfyUI, the runtime and models, with offline install support.

Use cases

  • Generating short AI videos on a home PC with an 8 GB to 16 GB GPU.
  • Running MiniMax H3 locally on a Mac with Apple silicon as a preview.
  • Adding video generation to an existing ComfyUI pipeline with LoRAs.
  • Hosting a generation server on a Linux machine and accessing it over the network.

Quick start

  1. 1

    Clone the repository and open its folder

    git clone https://github.com/FlashML-org/FreeVideo.git && cd FreeVideo
  2. 2

    Run the setup script, which installs FreeVideo, ComfyUI and the models, then opens it in the browser

    ./setup.sh

Things to know

  • The model weights use the MiniMax H3 Community License, which has territorial and acceptable-use restrictions, even though the code is Apache-2.0.
  • The macOS build is a preview for Apple silicon only and is not notarized by Apple, so macOS blocks it on first open.
  • Full offline installation requires several large model packs downloaded separately.

In collections

Topics

FAQ

What is FreeVideo?
Local video generation app and ComfyUI plugin that runs MiniMax H3 on consumer GPUs with as little as 8 GB of VRAM. FreeVideo is a local inference engine for the MiniMax H3 video model, built on the OpenVDN 8-step VDN-H3 model. It targets users who want to generate videos on their own consumer GPU rather than through a hosted service. It coordinates VRAM, system memory and disk by streaming weights, prefetching them asynchronously and computing in chunks, so generation can run with as little as 8 GB of VRAM and 16 GB of RAM.
Is FreeVideo free to use? What license does it have?
FreeVideo is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is FreeVideo still maintained?
Actively maintained. The last commit to FreeVideo was on Nov 6, 2026.
What can I use FreeVideo for?
Generating short AI videos on a home PC with an 8 GB to 16 GB GPU. Running MiniMax H3 locally on a Mac with Apple silicon as a preview. Adding video generation to an existing ComfyUI pipeline with LoRAs. Hosting a generation server on a Linux machine and accessing it over the network.
How do I install FreeVideo?
Clone the repository and open its folder: git clone https://github.com/FlashML-org/FreeVideo.git && cd FreeVideo Run the setup script, which installs FreeVideo, ComfyUI and the models, then opens it in the browser: ./setup.sh
Where can I get FreeVideo?
The source code and installation instructions are on GitHub: https://github.com/FlashML-org/FreeVideo

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects